Hybrid Java Compilation and
Optimization for Digital TV
Dong-Heon Jung, Hyeong-Seok Oh,
Soo-Mook Moon
smoon@snu.ac.kr
School of EECS
Seoul National University, Korea
Accelerating DTV S/W Platform
 DTV allows data-broadcasting
• Sending data as well as picture/sound
 Data-broadcasting platform is based on Java
• Java xlets + Java middleware at the set-top box
 Java is slow, so use just-in-time compilation (JITC)
 Propose using ahead-of-time and idle-time
compilation/optimization as well
• Hybrid compilation and optimization
Microprocessor Architecture & System Software Lab
2
Executing xlet with JITC only
Microprocessor Architecture & System Software Lab
3
Executing xlet with Hybrid
Microprocessor Architecture & System Software Lab
4
Outline
 Background on digital TV S/W platform
• Xlet lifecycle
• DTV acceleration
 Hybrid Java Compilation and Optimization
• JITC for xlet methods
• AOTC for system/middleware methods
• ITC and ITO for xlets
 Experimental Results
 Summary
Microprocessor Architecture & System Software Lab
5
Digital Television (DTV)
 DTV sends digital signals instead of analog signals
• Higher definition pictures and clearer sounds
 Remaining bandwidth can be used for sending data
• General information: traffic, weather, news, stock, …
• Program-specific information (plot, cast, director,…)
• Interaction using a return channel
– T-commerce, T-banking,
T-government, …
 Provides the data-broadcasting, interactive TV (iTV)
Microprocessor Architecture & System Software Lab
6
Java for Interactive TV
 One key technology for iTV is Java
• Many open standards are based on Java
– DVB-MHP (satellite), OCAP (cable), ACAP (terrestrial)
 Programmed using xlet applications
• xlet classes + image/text files
• Downloaded to the DTV set-top box
• Interact with middleware/system classes at the set-top
 xlet execution starts only when the user initiates it
Microprocessor Architecture & System Software Lab
7
Sending and Receiving xlet App.
 Xlet application is sent via carousel mechanism
• Send a stream of xlet files
repeteatedly in a round-robin
• Carousel file manager in the
set-top handles the receiving
 When the DTV is turned on,
• JVM starts and the application manager starts
• Then xlet application for current channel start its lifecycle
Microprocessor Architecture & System Software Lab
8
The xlet Lifecycle
Not Loaded
Loaded
When starting download of xlet application
When loading xlet’s main class file
initXlet()
Paused
startXlet()
destroyXlet()
Destroyed
pauseXlet()
When switching to a different channel
Started
At Started state, a red-dot appears on the TV screen
Microprocessor Architecture & System Software Lab
9
An Example of xlet Execution
(a) Display Red-dot
(c) Select xlet menu
(b) Display xlet Menu
(d) Display Slected Menu
Microprocessor Architecture & System Software Lab
10
DTV Java Architecture
 Two types of classes in DTV Java Platform
• System/middleware classes statically installed at DTV
• xlet classes dynamically downloaded from TV station
 Similarities in other platforms
• Mobile phone Java platform: MIDP middleware + midlet
• Bluray disk Java platform: BD-J middleware + xlet
 Both class types are getting more substantial
• E.g., MIDP -> JTWI -> MSA
 How to accelerate these substantial, dualcomponent Java platforms?
Microprocessor Architecture & System Software Lab
11
Hybrid Compilation and Optimization
 Current wisdom of Java acceleration: JITC
• Compile bytecode to machine code at runtime
• In DTV, do JITC both xlets and system/middleware
 Our proposal: hybrid compilation and optimization
• Ahead-of-time compilation (AOTC) for system/middleware
• Idle-time compilation (ITC) for xlets
• Idle-time optimization (ITO) for images and text fonts
Microprocessor Architecture & System Software Lab
12
Hybrid Environment for DTV
XLET Applications
Set-top Box
Object Carousel File Manager
Phone Me Advanced
Middleware
& system
methods
Xlet
methods
Xlet images
and texts
AOTC
JITC/ITC
ITO
Persistent Storage
OS & Hardware
 We actually built a hybrid environment for a DTV
based on a PhoneME Advanced (CDC) VM
Microprocessor Architecture & System Software Lab
13
AOTC for System/Middleware
 Employ AOT module in PhoneME Advanced VM
• Compile pre-chosen methods using JITC and save in a file
• When JVM starts officially, use the machine code directly
– With no interpretation or compilation overhead
 Two issues
• Which methods to AOTC in system/middleware?
– AOTC only those methods compiled at least once by JITC
• Optimization
– AOT-generated code is worse than JITC-generated code
Microprocessor Architecture & System Software Lab
14
AOT Enhancements
 AOT inlining without runtime behavior
• Implement inlining based on profile-feedback
 No code patch optimization
• Translated code for class initialization check, GC-check
can be patched
 Relocation prohibits some optimizations
• Constant pointer optimization
Microprocessor Architecture & System Software Lab
15
Idle-Time Compilation (ITC) for xlet
 Compile xlet methods in advance (idle-time)
• Saves the JITC and interpretation overhead
• Use our enhanced AOT
• Assign a separate, lowest-priority thread for ITC to
reduce the delay of the main thread (displaying red-dot)
• OK even if user executes xlet in the middle of ITC
Microprocessor Architecture & System Software Lab
16
Idle-Time Optimization for Images
 Loading/decoding of xlet images occur at runtime
• Just-in-time when they are needed
• Their overhead is substantial, taking much of running time
 Propose pre-loading/decoding during idle-time
 Two issues
• When we start pre-loading/decoding in the xlet lifecycle
– Started state or Not-loaded state: Do not work
– Loaded state: good
• How we perform pre-loading/decoding transparently
– Use the ITC thread
 Useful even when user executes xlets early and becomes idle
Microprocessor Architecture & System Software Lab
17
Just-in-Time Loading/Decoding
Start
Run java code of
selected menu
Initialize Xlet
Request Image
Object
Image is
cached??
Start Xlet
Display red-dot
no
yes
User select the menu
Display selected menu
Load image
from cache
Perform Image
loading/decoding
Save the image to
image cache
Finish xlet
Get image object
End
Finish java code of
selected menu
Microprocessor Architecture & System Software Lab
18
Pre-loading/decoding
Start
Start Imagepreprocessing thread
Initialize Xlet
New file is
received?
yes
Get
each image file name
Is preprocessed?
no
Perform Preloading/decoding
no
Terminate
the thread
Start Xlet
Display red-dot
User select the menu
Display selected menu
yes
Finish xlet
End
Save the image to
cache
Microprocessor Architecture & System Software Lab
19
Idle-Time Optimization for Texts
 Creating some font objects occur at runtime
 Pre-creating of them at idle-time
Microprocessor Architecture & System Software Lab
20
Experimental Results
 Experimented on a commercial DTV platform with
real, on-air xlets broadcasted in Korea
 Experimental Environment
•
•
•
•
DTV set-top box 333MHZ MIPS CPU with 128MB memory
Linux with kernel 2.6
Sun’s phoneMe Advanced MR2 version
Advanced common application platform (ACAP)
Microprocessor Architecture & System Software Lab
21
Benchmarks
 xlets of three terrestrial TV stations in Korea
•
•
•
•
Designated by A, B, C
News, weather, traffic, and stock menu items
Interested in running time of each menu item
Size of xlet applications (KB)
Station
A
Station
B
Station
C
class
image
text & et
c.
Total
276
1,348
344
1,968
360
1,596
372
2,328
448
1,280
288
2,016
Microprocessor Architecture & System Software Lab
22
Distribution of Method Calls
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
NEWS
WEATHER TRAFFIC
STOCK
Station A
xlet method
NEWS
WEATHER TRAFFIC
STOCK
Station B
system method
NEWS
WEATHER Geomean
Station C
middleware method
Microprocessor Architecture & System Software Lab
23
Distribution of JITCed Methods
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
NEWS
WEATHER TRAFFIC
STOCK
Station A
xlet method
NEWS
WEATHER TRAFFIC
STOCK
Station B
system method
NEWS
WEATHER Geomean
Station C
middleware method
Microprocessor Architecture & System Software Lab
24
Image Loading/Decoding Overhead
100%
80%
60%
40%
20%
0%
NEWS
WEATHER TRAFFIC
STOCK
NEWS
WEATHER TRAFFIC
Station A
Image processing runtime portion
Station B
STOCK
NEWS
WEATHER Geomean
Station C
others (java & native code)
Microprocessor Architecture & System Software Lab
25
Running Time Impact of AOTC
18,000
Running time (ms)
16,000
14,000
12,000
10,000
8,000
6,000
4,000
2,000
Station A
JITC only
Station B
JITC + AOT(original)
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
0
Station C
JITC + AOT(enhanced)
Microprocessor Architecture & System Software Lab
26
Performance Impact of AOTC
180%
Speedup
160%
140%
120%
100%
80%
60%
40%
20%
Station A
JITC only
Station B
JITC + AOT(original)
Geomean
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
0%
Station C
JITC + AOT(enhanced)
Microprocessor Architecture & System Software Lab
27
Impact of Pre-loading/decoding
18,000
Running time (ms)
16,000
14,000
12,000
10,000
8,000
6,000
4,000
2,000
Station A
JITC only
Station B
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
0
Station C
JITC + image pre-loading/decoding
Microprocessor Architecture & System Software Lab
28
Impact of Text Font Pre-creation
18,000
Running time (ms)
16,000
14,000
12,000
10,000
8,000
6,000
4,000
2,000
Station A
JITC only
Station B
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
STOCK
TRAFFIC
WEATHER
NEWS
0
Station C
JITC + Font Pre-creation
Microprocessor Architecture & System Software Lab
29
Overall Running Time of Hybrid
18,000
An average of 150% reduction
(15% by AOTC)
16,000
14,000
12,000
10,000
8,000
6,000
4,000
2,000
0
NEWS
WEATHER TRAFFIC
Station A
JITC only
STOCK
NEWS
WEATHER TRAFFIC
STOCK
Station B
NEWS
WEATHER
Station C
JITC + AOT(enhanced) + image/text pre-processing
Microprocessor Architecture & System Software Lab
30
Impact on Transparency
Running time (ms)
90,000
80,000
70,000
60,000
50,000
40,000
30,000
20,000
10,000
0
JITC only
Our Optimized
VM
Station A
JITC only
Our Optimized
VM
Station B
Red-dot
JITC only
Our Optimized
VM
Station C
Pre-processing completion
Microprocessor Architecture & System Software Lab
31
Summary and Future Work
 Proposed hybrid compilation/optimization for DTV
• Just-in-time, ahead-of-time, and idle-time
• Improves performance dramatically than JITC-only
– With little change to other DTV behavior
• Some ideas would work for other dual-component Java
 Some future work
• AOTC for system/middleware beyond AOT
– By performing off-line AOTC with full optimizations enabled
The idea of pre-loading/decoding has been filed for patent application.
Microprocessor Architecture & System Software Lab
32
Thank you!